Table of Contents

flink Linux Kernel Modules

This is the user documentation for the flink kernel modules. They offer drivers capabilities to communicate with various hardware interfaces. For more information about the inner workings see flink Linux Kernel Modules on Github. The driver API can be found under API

Overview

When flink is used on a Linux based system the flink library in userspace communicates with the hardware by means of several kernel modules. The basic module is flink.ko. Further, every flink device needs one more kernel module for each hardware interface which is used to communicate with an external FPGA. This could be flink_pci.ko, flink_spi.ko, flink_lpb, flink_axi or flink_eim. They use the appropriate underlying subsystem for the particular bus.

Requirements

Building

Generating the documentation

Create the makefile with cmake (in the root directory):

cmake .

After this generate the documentation as given in the Doxyfile by :

make doc

Installation

Module Parameters

Determine the availabe parameters for each kernel module with the command modinfo. The following table shows the currently available modules with their parameters.

module parameters description example
flink.ko core module
flink_pci.ko vendor id, product id for pci interface flink_pci vid=0x1172 (Altera) pid=4
flink_spi.ko device memory length (see below) for spi interface flink_spi dev_mem_length=0x280
mpc5200/flink_lpb.ko device memory length (see below) for local plus bus mpc5200/flink_lpb dev_mem_length=0x280
imx6/flink_eim.ko for EIM bus
zynq/flink_axi.ko either hardcoded into module or reading device tree node for AXI bus

Busses like PCI have discoverability built into them. Any device connected to such a bus can tell the system, what kind of device it is and what resources it uses. On the SPI this is not possible. Therefore, you must provide the total device memory length when loading the module. This length is determined by the number and types of the subdevices implemented in VHDL, see flink VHDL.
If an info subdevice is present, it will give the information about the total device size and the parameter can be ommitted, see Info Subdevice.

Performance

When parallel busses such as PCI or LocalPlusBus are used, any flink transfer happens very quickly and basically depends on the speed of the bus and the FPGA. Care has to be taken when using the SPI interface. The standard SPI subsystem in a standard Linux distribution uses an internal queueing mechanism which is rather slow. Subsequent transfers of a message are separated by approximately 70μs. Further, every flink read or write includes a 8 byte transfer (address and data phase). Though SPI allows for full duplex transfers this feature cannot be used here.
IMPORTANT When using SPI timing considerations become crucial. As a rule of thumb: Every flink transfer (such as updating a PWM high time or reading a encoder value takes up to 200μs. SPI should be used for low-frequency control application with not too many parallel channels.